Pathway to net zero: building AI into climate risk strategies

Can generative AI (GenAI) or machine learning both help set and effectively deliver net zero strategy? Is it, indeed, the tool that will change the face of ESG and sustainability disclosure and strategy? Is the use of GenAI as a tool, the silver bullet, a way to help move a company (or a country) to net zero, efficiently, and at speed? In principle, the use of GenAI as an innovative tool can help companies (or anyone with a net zero strategy) work ‘smarter not harder’ and achieve net zero and other climate change goals. But, all the externalities of using the tool must be understood, both positive and negative. 

Climate risk and climate strategy 

Climate risk is steadily moving the needle from awareness through issue and priority to strategy. The evolution has been subtle until just a few years ago, where the leap from awareness to a priority began to bite. This was in concert with rising corporate discussions around ‘resilience’, ‘adaptation’ and ‘mitigation’. The business case to address climate change impacts and all related risks is urgent. Past strategies combatting climate risks no longer work. There is a real need to be creative, innovative, and ready to look beyond the traditional. Climate risks remain the same, but the approach to measure, monitor and mitigate the risk continues to challenge companies, countries and communities. 

These types of climate risks are familiar to any student of climate change, or any business which has been impacted by climate events. As the understanding around climate risks moves to becoming more strategic, clearly there are business opportunities for any company from being merely resilient to the proactive seeking of innovative solutions. 

Climate risks are couched in four distinct and familiar categories: liability risk, reputational risk, transitional risk and physical risk. The key is for corporations to be clear and transparent about their goals and efforts to tackle such risks. The opposite of transparency is seen as ‘greenwashing’. These risks require specific and robust data, to ensure that the risks are measured, monitored and managed; all key elements underpinning a robust and credible strategy. The gathering of that data feeds directly into goals and target delivery, as well as supporting a well-informed business strategy - a strategy that includes all business risks, including climate risk.  

Companies require clarity around their climate risks. It is just as important to approach those risks with a view towards resilience opportunities and creativity. There is plenty of research outlining how a company can address climate risks; from alternative energy sources to decarbonisation strategies, from investment in new markets to clear business opportunities as solution providers.  

Net zero relies on a robust understanding of climate risk and is the next step in effectively moving the needle from a priority to concise strategy. The question is what contributions can AI make to this understanding?  

AI, climate and energy  

AI has often been described as a transformative general-purpose technology, analogous to electricity (i.e. a blog by AI luminary Andrew Ng [1]) or, (in this context) ironically as “the new oil” [2]; but as exciting as these categorisations are, it may be more useful to identify specific properties and impacts of AI relevant to net zero strategy.  

• AI to provide insight: one of the early relatively successful applications of GenAI has been to use it to create summaries of large and complicated documents. For some time before GenAI became all the rage, machine learning systems were being effectively used to mine for information in structured data, providing executives with actionable insights on their business operations. Whilst it is clear that the new approach does not provide an oracle that can be relied on to pull comprehensive and reliable reports out of a company’s document store, GenAI does appear to be a more accessible and robust method for finding insight in complex unstructured data. 
• AI to provide predictions: accurate and reliable predictions of energy consumption can be used to create efficient plans that avoid carbon intensive energy sources. Additionally, prediction can be used to create simulations that validate energy saving interventions. One of the first applications of the new wave of AI by Google was the development of models of energy use in Google datacentres. Reportedly, this application reduced the amount of energy used for datacentre cooling by 40% [3]. More recently, DeepMind has developed medium range weather forecasting models that are reportedly more accurate than previous supercomputer simulations and consume 1/1000th of the energy to produce a forecast. The technology underpinning this use case is not GenAI in the strictest sense, but it is part of the emergent technology of AI created by the boom in data and compute in the last 10 or so years.  
• AI as a more efficient engine: the use of AI for weather forecasting is an interesting early indicator of a surprising potential net zero benefit. In the next section, we will highlight the very high energy consumption of the current generation of AI models; however, there is potential for the implementation of AI-powered systems that are more energy-efficient and less carbon-intensive than traditional business processes and IT systems.  

There are some specific ways that AI could support an effective net zero strategy; creating insight to drive initiatives, making predictions that build resilience in plans and enable efficiency, and as a direct driver of energy efficiency. Unfortunately, there is another side of AI as a technology that business leaders must be cognisant of and that should be used as way to challenge work using AI in the three ways described above. The boom in AI is a direct driver of carbon emissions and other environmental damage. The International Energy Agency (IEA) has released a prediction [4] that AI will consume 10x more electricity in 10 years’ time. In the report, the IEA estimates that a Google search requires 0.3Wh of electricity, whilst an AI model like ChatGPT requires 2.9Wh. 

Net zero strategy and AI: friend or foe?  

If modern AI can be used in an efficient and targeted way, and if it can be effectively fed the data that it needs, it may come into its own. Specifically, the ability of AI systems to make useful predictions about complex scenarios with ‘fat tails’ in their distributions (such as weather and climate) and the development of simulations that can be used to demonstrate the utility of alternative policies and interventions may provide decision makers with new high utility options for getting to net zero. 

This may be the way that innovative technologies will have a lasting impact on the ESG agenda, identifying clear pathways to managing climate risks. It can be used to work ‘smarter not harder’, but those who use AI in the quest to be net zero, must also be cognisant of the energy intensity imbedded in the ‘smarter’ working strategy. GenAI is not a silver bullet, but a tool that should be used responsibly and with a clear understanding and accounting for the positive and negative impacts. 

This blog is co-authored by Simon Thompson (Head of AI, ML & Data Science, GFT) and Dr Tauni Lanier (Sustainability and ESG Director, BDO).

[1] Andrew Ng’s blog “AI is the new electricity”. https://www.gsb.stanford.edu/insights/andrew-ng-why-ai-new-electricity 

[2] Department of Defence calls AI “the new oil”. https://www.defense.gov/News/News-Stories/Article/Article/2386956/defense-official-calls-artificial-intelligence-the-new-oil/ 

[3] Deepmind reduces Google Data Centre energy used for cooling by 40%. https://deepmind.google/discover/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-by-40/  

[4] Electricity 2024 (IEA report) https://iea.blob.core.windows.net/assets/ddd078a8-422b-44a9-a668-52355f24133b/Electricity2024-Analysisandforecastto2026.pdf